Molecular Dynamics Study of Polarization Effects on AgI - The Journal

Jan 19, 2008 - Three different models of AgI are studied by molecular dynamics simulations. The first one is the rigid ion model (RIM) with the effect...
31 downloads 10 Views 181KB Size
Download PDF
1718

J. Phys. Chem. B 2008, 112, 1718-1728

Molecular Dynamics Study of Polarization Effects on AgI Vicente Bitria´ n and Joaquim Trulla` s* Departament de Fı´sica i Enginyeria Nuclear, UniVersitat Polite` cnica de Catalunya, Campus Nord UPC, 08034 Barcelona, Spain ReceiVed: September 21, 2007; In Final Form: NoVember 7, 2007

Three different models of AgI are studied by molecular dynamics simulations. The first one is the rigid ion model (RIM) with the effective pair potential of the Vashishta and Rahman form and the parametrization proposed by Shimojo and Kobayashi. The other two are polarizable ion models in which the induced polarization effects have been added to the RIM effective pair potential. In one of them (PIM1), only the anions are assumed to be polarizable by the local electric field. In the other one (PIM2s), the silver polarization is also included, and a short-range overlap-induced polarization opposes the electrically induced dipole moments. This short-range polarization is proved to be necessary to avoid overpolarization when both species are assumed to be polarizable. The three models reproduce the superionic character of R-AgI at 573 K and the liquid behavior of molten AgI at 923 K. The averaged spatial distribution of the cations in the R-phase obtained for PIM1 appears to be in better agreement with experimental data analysis. The PIM1 also reproduces the structure factor prepeak at about 1 Å-1 observed from neutron diffraction data of molten AgI. The three models retain in the liquid phase the superionic character of R-AgI, as the mobility of the cations is significantly larger than that for the anions. The ionic conductivity for the polarizable ion models is in better agreement with experimental data for R-AgI and molten AgI.

1. Introduction Silver iodide (AgI) is one of the most extensively studied superionic conductors, namely, systems characterized by having exceptionally high (liquid-like) values of ionic conductivity σ in the solid state. This high conductivity is reached because of the diffusive motion of one of the constituent species through the sublattice formed by the particles of other species.1 At room temperature and pressure, AgI consists of a mixture of the γ and β phases, which have the cubic zinc-blende and the hexagonal wurtzite structure, respectively.2 The β-phase becomes the more stable phase above 410 K, and a first-order structural phase transition to the superionic R-phase takes place at 420 K accompanied by an increase in σ of around 3 orders of magnitude. In the R-phase, the iodides form a body-centered cubic (bcc) lattice, while the silver ions diffuse through it. It has been shown experimentally, by means of neutron3-5 and X-ray3,6 diffraction and extended X-ray absorption fine structure studies,7 that the Ag+ ions predominantly occupy the tetrahedral sites located on the faces of the bcc unit cells, with the diffusion between these sites occurring primarily along the [110] directions via trigonal sites. The high R-AgI ionic conductivity is related to this high degree of disorder in the Ag+ positions (there are 12 available tetrahedral sites and only 2 cations in each unit cell). The conductivity at the phase transition temperature (420 K) is σ ) 1.3 (Ωcm)-1, a value comparable to that of molten salts, it increases with temperature to reach σ ) 2.6 (Ωcm)-1 at the melting point (825 K), and decreases by approximately 10% upon melting.8 Computer simulation has been a powerful tool in the study of AgI properties, both in solid and molten phases. In the first molecular dynamics (MD) study, Vashishta and Rahman9 introduced a polynomial form for the effective pair interionic * Corresponding author. E-mail: [email protected].

potential parametrized using crystal data. They obtained good agreement with experiment for the calculated properties of R-AgI. Later, Parrinello, Rahman and Vashishta10 carried out MD simulations in which the volume and shape of the simulating cell were allowed to change with time. With the same functional form but a different parametrization from that in ref 9, their effective pair potential (which we will refer to as PRV potential) was capable of describing the β T R transition. The PRV potential was also used by Tallon11 to simulate a phase diagram of silver iodide similar to that observed for actual AgI, by Chiarotti et al.12 to study the motion of Ag+ ions in the R-phase, and by Madden et al.,13 who suggested the ordering tendency of Ag+ ions observed in simulation as the mechanism responsible for the R f β transition. The PRV potential has also been used in simulations of the molten phase. Howe et al.,14 as a result of a diffuse neutron scattering experiment, noted the similarity between the structure of R-AgI and that of molten CuCl, which melts from a superionic phase. This result prompted Stafford and Silbert15 to suggest that, since the PRV potential was suitable for a description of the superionic R-phase, it should also be a good potential choice to describe molten AgI. They carried out theoretical calculations of the radial distribution functions and structure factors within the hypernetted chain approximation (HNC). Shortly after this work, Takahashi et al.16 published the first neutron diffraction data for the coherent static structure factor of molten AgI and found qualitative agreement between theoretical and experimental results. Moreover, MD simulations using the PRV potential suggested that molten AgI, near melting, somehow retains the superionic character of the R-phase.17,18 Later, Shimojo and Kobayashi19 slightly modified the parametrization of the PRV potential in order to reproduce the R-phase features at the appropriate experimental density (larger than that in ref 10).

10.1021/jp077604a CCC: $40.75 © 2008 American Chemical Society Published on Web 01/19/2008

Polarization Effects on AgI The ionic induced polarization was not taken into account in the above simulation works. However, Madden and Wilson20 showed the important role of polarization in some ionic systems, such as alkaline earth halides, and stated that it can account for phenomena often attributed to a certain degree of covalency. They proposed two different mechanisms for the induction of ionic polarization: a purely electrostatic induction governed by a fixed polarizability, and a mechanical short-range polarization that depends on the nearby ions. This latter contribution, which opposes the former, allows for the fact that the dipole moments created by the field are damped at short interionic separations where overlap effects become significant. The influence of the induced polarization on the properties of other silver halides such as AgCl and AgBr has already been investigated. Wilson et al.21 simulated solid and liquid AgCl by using a model in which dipole and quadrupole polarization effects were added to a pair potential of the Born-Mayer form. Their results for the optic phonon dispersion curves and the melting temperature were in much better agreement with experiment than those obtained with a simple pair potential, and they qualitatively reproduced the distinctive three-peak feature in the main broad peak of the experimental structure factor. Later, Trulla`s et al.22 showed that the three-peak structure of molten AgCl can also be reproduced if only the anionic induced dipole polarization contributions are added to a pair potential of the form proposed by Vashishta and Rahman in ref 9. Furthermore, they found that, unlike ref 21, the results for the conductivity were in good agreement with experimental values. Recently, we carried out an MD study on molten AgBr by using a rigid ion model and a polarizable ion model in which only the anionic induced polarization is considered.23 The latter reproduces the broad main peak of the structure factor obtained by neutron diffraction experiments, although the smoothed threepeak feature of this broad peak is slightly overestimated. We also studied a model in which both species were polarizable, but a non-physical overpolarization (“polarization catastrophe”) took place. In order to solve this problem, according to refs 20 and 21, we added a short-range damping contribution to the dipole moment that opposes the dipole induced by the electric field created by the ionic charges. However, the polarization catastrophe was not avoided. We saw that simulations could be stable longer than 105 time steps, but for some improbable but possible critical configurations, two unlike ions became overpolarized. The purpose of this work is twofold. First, we propose a new model that avoids the polarization catastrophe when both species are polarizable. In this model, the short-range damping contribution opposes the dipole induced by the electric field created by both the ionic charges and the induced dipoles. The second aim is to complete the paper series about the polarization effects on the structure and ionic transport properties of molten silver halides. We have already studied molten AgCl and AgBr in previous papers,22,23 and we have recently published a preliminary MD study on the structure of molten AgI and the origin of the prepeak observed in the experimental structure factor.24 In this recent communication, we used a polarizable ion model in which the anions are only polarizable by the electric field. Now, in the present work, we have also carried out MD simulations of a new model in which both cations and anions are polarizable. Furthermore, we have also checked that the ionic models reproduce some well-known characteristics of the solid R-phase. The layout of the paper is as follows. We describe the models in section 2. In section 3 we describe the computational details.

J. Phys. Chem. B, Vol. 112, No. 6, 2008 1719 TABLE 1: Potential Parameters of the RIM (Eq 2) for AgI, with |za| ) 0.5815 Ag+-Ag+ ηab Hab/eVÅηab Pab/eVÅ4 a Cab/eVÅ6 a

Ag+-I-

I--I-

9 1310 14.9 0

7 5328 29.8 84.5

11 0.16 0 0

In the polarizable ion models, Pab ) 0 for all interactions.

In section 4 we present and discuss the results of our simulations. Finally, we summarize our results in the concluding remarks of section 5. 2. Interaction Models 2.1. The Rigid Ion Model. The potential energy of the rigid ion model (hereafter referred as RIM) is N-1 N

URIM )

φVR ∑ ∑ ij (rij) i)1 j>i

(1)

where N is the number of ions, and rij is the distance between two ions. The functional form of the effective pair potential is that originally proposed by Vashishta and Rahman,9 which has been widely used to describe silver and copper halides (for a review, see ref 1), 0 φVR ab (r) ) φab(r) -

Pab r4

(2)

with

φ0ab(r)

zazbe2 Hab Cab + η - 6 ) r r ab r

(3)

where we use the a and b subscripts to denote species Ag+ or I- rather than particles. The first term on the right-hand side of eq 3 is the Coulomb interaction between ionic charges, with |za| < 1 the effective charge in units of the fundamental charge e; the second models the repulsion between the ions arising from the overlap of the outer shells of electrons; and the third is the van der Waals contribution. From recent ab initio MD simulations of molten AgI, Shimojo et al.25 found that the valence electrons of Ag ions are not completely transferred to I ions, which supports the use of effective charges |za| < 1. The last term in eq 2 is the effective monopole-induced dipole attractive interaction, with Pab ) (Razb2 + Rbza2)e2/2 where Ra are the electronic polarizabilities. This is the way in which polarization terms are approximated in the RIM as two-body terms, ignoring the many-body nature of induced polarization. We have used the parametrization proposed by Shimojo and Kobayashi (see Table 1).19 At this point it is worth noting that, in the original PRV potential, Hab ) A(σa + σb)ηab, where σAg ) 0.53 Å and σI ) 2.20 Å are related to the ionic radii.10 Although Shimojo and Kobayashi slightly modified the values of Hab, their parametrization keeps the large size difference between cations and anions. 2.2. The Polarizable Ion Models. The polarizable ion models are constructed by adding the many-body induced polarization interactions to the pair potential φ0ab(r) of eq 3 with the same parameter values given in Table 1. We assume that two types of point dipoles are induced in an ion placed at position ri. The first one is the dipole induced by the local electric field Ei due to all other ions, whose moment is given by the polarizability Ri in the linear approximation RiEi. The second type is the

1720 J. Phys. Chem. B, Vol. 112, No. 6, 2008

Bitria´n and Trulla`s

deformation dipole induced by short-range overlap effects due to the neighboring ions.26 It can be physically interpreted as a correction for the fact that the electronic distribution is not well represented by point charges and dipoles at small distances.27 For the resulting dipole moment pi we use the following constitutive relation:

pi ) RiEi + psi

(4)

N-1 N

UPIM )

1

N

Ei )

+

Epi

)

∑ j*i

N

Eqij

+

Epij ∑ j*i

(5)

where

Eqij )

zje rij

3

rij and Epij ) 3

pj‚rij rij

5

rij -

pj rij3

(6)

are the electric field at ri created by the charge qj ) zje and by the induced dipole on the ion j (within the point dipole approximation), respectively. If psi is approximated to be additive, a general constitutive relation for psi can be written as N

psi ) -Ri

[f q(rij)Eqij + f p(rij)Epij] ∑ j*i

(7)

where f q(r) and f p(r) are suitable damping functions only effective over length scales corresponding to the nearestneighbor separation and limiting value 1 as r f 0. By choosing this form, we ensure that the short-range terms cancel the dipole moment induced by the electric field if two ions approach unphysical separations. Notice that eq 7 could have been written in a more general form with f qij(r) and f pij(r) instead of f q(r) and f p(r), but this is not the case in this work. The relation suggested by Madden and co-workers21,26 and used by us in ref 23, is recovered from eq 7 with f p(r) ) 0. This relation works well for a variety of divalent and trivalent metal halide models in which only the anions are assumed to be polarizable.28 However, we found that the polarization catastrophe takes place for silver halide models in which both species are polarizable. This is the reason why, in this work, we also include the damping of the dipole moment induced by the Epij terms, in a similar way as other authors (see, for example, refs 29 and 30). In this work, we have chosen the simplest extension of the model used in ref 23 by assuming

f (r) ) f (r) ) f(r) ) exp(-r/F) q

p

4

(r/F)k

k)0

k!

∑

(8)

where f(r) is the Tang and Toennies dispersion damping function.26,31 The parameter F determines the length scale over which the damping acts, but we will characterize f(r) by λ ) 4.671F, the distance at which f(λ) ) 0.5. The short-range polarization effects are eliminated in the limit λ ) 0. Substitution of eqs 5-8 in eq 4 leads to a system of linear equations that uniquely determines the N dipole moments {pi} for given ionic positions {ri}. The general expression of the potential energy for the above polarizable ion models can be written as

N

∑ 2 i)1

pi‚Epi +

N

1

pi2

∑ 2 i)1 R

+ Us (9)

i

where N

where the superscript “s” stands for “short-range damping”. The local electric field at ri is

Eqi

N

pi‚Eqi ∑ ∑ φ0ij(rij) - ∑ i)1 j>i i)1

Us )

N

∑ ∑ i)1 j*i pi‚

f q(rij)Eqij +

1

N

N

f p(rij)Epij ∑ ∑ j*i

2 i)1

pi‚

(10)

The first term on the right-hand side of eq 9 corresponds to the potential energy of a rigid ion model whose ions interact via the pair potential φ0ij(r). The second, third, and fourth terms give the usual potential energy of point dipoles linearly induced by the electric field.27,32,33 The last term is due to the inclusion of the short-range damping polarization. It is obtained by imposing that, given an ionic configuration, the minimum of the potential energy occurs when the values of the dipole moments satisfy eq 4, i.e., the gradient of UPIM with respect to each dipole moment vanishes. The fact that ∇piUPIM ) 0 (for i ) 1,..., N) implies that all contributions to the forces coming from terms involving derivatives with respect to the dipole moments are zero. Then, only the explicit spatial derivatives of UPIM have to be considered, and the force acting on the ion i can be written as in ref 22 plus an extra term due to the negative gradient of Us, whose expression is

N

Fsq i

)

∑ j*i

sp Fsi ) Fsq i + Fi

([

zje f

(rij)Epji

q

-

f˙ q(rij)

[

rij4

N

∑ j*i

)-

(

f

(rij)∇ri[pi‚Epij]

p

]

(pi‚rij)rij -

zie f q(rij)Epij -

Fsp i

(11)

+

f˙ q(rij) rij4 f˙ p(rij) rij

])

(pj‚rij)rij

)

(pi‚Epij)rij

(12)

(13)

where f˙ q(r) ) df q(r)/dr and f˙ p(r) ) df p(r)/dr. Equation 12 is equivalent to the extra term given in ref 23, and eq 13 corresponds to the new short-range damping contributions introduced in this work. In this work we present the results of two polarizable ion models. In both models, the anions are assumed to be polarizable with RI ) 6.12 Å3 (ref 19). In the first one, which we denote PIM1, the cations are not polarizable (RAg ) 0), and the dipole moments of the anions are induced only by the local electric field (λ ) 0). Then f(r) ) 0, and the terms related to the damping effects, i.e., psi in eq 4, Us in eq 9, and the force contribution given by eq 11, vanish. In the second model, which we denote PIM2s, the cation polarizability is RAg ) 1.67 Å3 (ref 34), and λ ) 1.84 Å. This λ value has been chosen according to the relationship λ ) (σAg+σI)/c proposed in ref 28, with the value of c derived from the λ used for AgCl in ref 21. 2.3. The Potential Energy of Two Isolated Ions. Substitution of eqs 5-8 in eq 4 leads to a system of linear equations that can be solved analytically for the two-particle case, and it is found that the potential energy of two isolated polarizable ions can be written as

Polarization Effects on AgI

J. Phys. Chem. B, Vol. 112, No. 6, 2008 1721

q 2 0 φPIM 12 (r) ) φ12(r) - [1 - f (r)]

(1/2)([R1z22 + R2z12]r3 - z1z2rc6[1 - f p(r)])e2 r(r3 + rc3[1 - f p(r)])(r3 - rc3[1 - f p(r)])

(14)

where rc ) (4R1R2)1/6. It is worth noting that eq 14 with R1 ) RAg ) 0, R2 ) RI * 0, and f q(r) ) f p(r) ) 0, as for PIM1, becomes equal to the effective pair potential φVR AgI(r) of eq 2 proposed for the RIM. Of special interest is the case of two unlike polarizable ions. In Figure 1 we plot φPIM 12 (r) for a cation and an anion with R1 ) RAg ) 1.67 Å3, R2 ) RI ) 6.12 Å3 and different damping functions. If f q(r) ) f p(r) ) 0, φPIM 12 (r) presents a divergence at rc ) (4R1R2)1/6 because the induced dipole moments become infinity at this polarization catastrophe distance. MD simulations of the whole system under these conditions polarize catastrophically whatever the initial configuration is. If f q(r) ) f(r) with λ ) 1.84 Å and f p(r) ) 0, as we did in ref 23 for AgBr, φPIM 12 (r) also diverges at the same critical distance rc, but now there is a maximum at a longer r. Therefore, the total force between two unlike ions that approach each other is repulsive at distances slightly larger than rc. MD simulations of the bulk under these conditions show that, for some improbable but possible configurations, an ion has a velocity large enough to overcome this energy maximum, whose form in each configuration depends significantly on the manybody effects. Although these simulations can be stable for a long time (more than 25 × 103 time steps), they also polarize catastrophically. However, if f q(r) ) f p(r) ) f(r) with λ ) 1.84 Å, as for PIM2s, φPIM 12 (r) does not exhibit any divergence. Hence, the conclusion drawn from this two-particle analysis, and confirmed by MD simulations of the whole system, is that the damping of the dipole moment contribution due the field created by the other dipoles must be considered in order to avoid the overpolarization. Given rc ) (4RAgRI)1/6 ) 1.86 Å, the value of λ must be larger than 1.6 Å, because, at lower values, the factor (r3 - rc3[1 - f p(r)]) in the denominator of eq 14 vanishes at a certain r, and the divergence is not avoided. In the case of two like polarizable ions, i.e., z1 ) z2 and R1 ) R2, the factor (r3 - rc3[1 - f p(r)]) in the denominator of eq 14 also appears in the numerator and cancels itself, and the potential energy for two ions of species a can be written as q 2 0 φPIM aa (r) ) φaa(r) - [1 - f (r)]

Raza2e2 r(r3 + 2Ra[1 - f p(r)]) (15)

without a polarization catastrophe distance even when f q(r) ) f p(r) ) 0 as for PIM1. However, MD simulations for a similar model of molten NaI with only the anions polarizable have shown that two anions can approach at distances at which they polarize catastrophically,35 even though this is not predicted for two isolated anions. Notice that eq 15 with R1 ) R2 ) RI * 0 and f q(r) ) f p(r) ) 0, as for PIM1, does not become equal to the effective pair potential φVR II (r) of eq 2. 3. Computational Details By using the potential models described above, we have simulated solid AgI at 573 K, the temperature at which Cava et al.4 carried out neutron diffraction measurements, and molten AgI at 923 K, the same temperature at which recent neutron diffraction data are available.36 The corresponding ionic number densities FN are 0.0306 ions/Å3 (ref 5) and 0.0281 ions/Å3 (ref 37), respectively.

Figure 1. Potential energy φPIM 12 (r) given by eq 14 for two isolated unlike polarizable ions with R1 ) RAg ) 1.67 Å3, R2 ) RI ) 6.12 Å3, the pair potential parameters of Table 1, and the following damping functions: f q(r) ) f p(r) ) 0 (dash-dotted line); f p(r) ) 0 and f q(r) ) f(r) with λ ) 1.84 Å (dashed line); and f q(r) ) f p(r) ) f(r) with λ ) 1.84 Å as for PIM2s (dotted line). The solid line corresponds to the potential energy of two unlike ions with R1 ) RAg ) 0, R2 ) RI ) 6.12 Å3, and f q(r) ) fp(r) ) 0, as for PIM1, which is equal to φVR AgI(r) given in eq 2.

MD simulations have been carried out with N ) 500 ions placed in a cubic box of side L with periodic boundary conditions, and using the Beeman’s algorithm with a time step of ∆t ) 5 × 10-15 s. The electric fields Eqi and Epi , and the corresponding long-range interactions between charges and induced dipole moments, have been evaluated by the Ewald method. The Ewald sums used in our simulations are given in ref 22. The Ewald parameters are the same as those used in ref 23. Notice that f q(rij)Eqij and f p(rij)Epij are short-range contributions and, thus, can be evaluated without the Ewald method. We have also carried out MD simulations of the RIM and PIM1 with 1000 ions, and we have verified that all the results are practically the same as those obtained with 500 ions. However, the time required in the PIM2s case with 1000 ions is 25 times longer than that for RIM (see below), beyond our present computing time capability. The system of equations that results from substituting eqs 5-8 into eq 4, whose solutions are the dipole moments for given ionic positions, may be solved by using the matrix inversion method. It is based on the matrix equation27,29

A‚p ) E′

(16)

where p ) (p1,...,pN) and E′ ) (E′1,...,E′N) are 3N-component vectors with E′i ) Σi*jEqij [1 - f q(rij)]. A is a 3N × 3N matrix that can be split into 3 × 3 submatrices Aij (i,j ) 1, ..., N) with elements

(Aij)µν ) δijδµν

1 - (1 - δij)[1 - f p(rij)](Tij)µν (17) Ri

where δij is the Kronecker delta, and Tij is the dipole field tensor,

Tij ) 3rijrij/rij5 - I/rij3

(18)

with I being the identity matrix. Inverting the matrix A we find, from p ) A-1‚E′, the dipoles for a given configuration. However, this exact method is very expensive computationally, and it is more convenient to calculate the dipole moments by using the prediction-correction iterative method proposed by Vesely.22,38 In this iterative method, an initial guess for the fields {Epi } and for the dipoles {pi} is made by using the dipole

1722 J. Phys. Chem. B, Vol. 112, No. 6, 2008 moments from the previous time step, and the dipole moments resulting from these fields and dipoles are evaluated using eq 4, which can be iterated to self-consistency. In the PIM1 case, we used the iterative method in all the time steps with a convergence limit |∆Epi |2/|Epi |2 < 10-4 for each ion. The convergence is reached after about 15 iterations, and the computer time spent is about 1.4 times longer than in the RIM case. In the PIM2s case, we found that the iterative process does not converge for a few percent of the time steps, as it leads to unstable oscillations in the predicted dipoles. Then, we used the iterative method when possible and the matrix inversion when the former does not work. With N ) 500, the iterative procedure converges after about 100 iterations in the 99.9% of the time steps, but the matrix inversion is needed for the other 0.1%, and the computer time spent is about 12 times longer than in the RIM case. With N ) 1000, the convergence is reached after about 300 iterations in 98% of the time steps, the matrix inversion is needed for the other 2%, and the computer time spent was about 25 times longer than in the RIM case, beyond our present computing time capability. All simulations were started by placing the anions on a bcc lattice and two cations per unit cell at two of the tetrahedral sites. Additionally, we checked that all the equilibrium properties obtained for the solid do not change if cations are placed in any other position of the anionic lattice. The solid properties were averaged over 500 × 103 equilibrium configurations. On the other hand, in order to get configurations in which the anions exhibit liquid behavior, we started the simulations of the molten phase at an ionic number density smaller and at a temperature higher than the density and temperature of interest. Once equilibrium was reached, we compressed and cooled the system in several steps, making sure that at each step the liquid behavior was preserved, until reaching the density and temperature of interest. Then, the liquid properties were averaged over 106 equilibrium configurations. This large number minimizes the statistical fluctuations in the structure factors and reduces the uncertainties in the values of the conductivity. The basic structural properties calculated in the simulations are the three partial radial distribution functions, gab(r), and the corresponding Ashcroft-Langreth structure factors, Sab(k).39,40 We have used the hybrid method described in ref 23 for the calculation of Sab(k). From Sab(k) and the coherent scattering lengths bAg ) 5.922 fm and bI ) 5.280 fm,41 the coherent static structure factor, normalized in such a way that it goes to 1 as k f ∞, is given by S(k) ) [bAg2SAgAg + bI2SII + 2bAgbISAgI]/ (bAg2 + bI2).42 Regarding the ionic transport properties, the self-diffusion coefficients, Da, and ionic conductivity, σ, have been calculated via the Kubo integral of velocity correlation functions and the Einstein relation for the slope of mean square displacements, in the same way as in ref 35. The estimated uncertainties for the Da values are less than 1%, and that for σ is less than 5%. The accordance between the values calculated using the Kubo and Einstein relations is within the error intervals. As in previous papers,22,23 the conductivity has been calculated using the formal charge |za| ) 1. Whereas in their interactions we assume the ions only “see” effective charges, in their transport we assume the ions carry with them the full complement of electrons. 4. Results 4.1. r-Phase Structure. In order to check that the three models reproduce the superionic behavior of the R-phase, we have carried out MD simulations at 573 K starting from a

Bitria´n and Trulla`s

Figure 2. Normalized silver ion density distribution in a (100) face of the conventional bcc cubic unit cell of R-AgI from MD simulations at 573 K using PIM1. The lengths represented on the axis are normalized by the lattice parameter a. There is an iodide site at each one of the vertices of the face. The sites at (0.5a, 0.25a), (0.25a, 0.5a), (0.5a, 0.75a), and (0.75a, 0.5a) are tetrahedral sites; those at the middle of the segment between two adjacent tetrahedral sites are trigonal sites; and those at the center of the face and at the middle of its four sides are octahedral sites.

configuration in which the iodides occupy bcc sites. In the three cases, the iodides remain around lattice sites, with a mean square displacement around 0.6 Å2. On the other hand, the silver ions show a liquid-like behavior independently of their initial positions and diffuse through the essentially rigid framework formed by the anions. A detailed analysis of the ionic transport properties is presented in subsection 4.3. In the three model simulations of R-AgI, the silver ions tend to be located on the faces of the cubic bcc unit cells most of the time. There is an iodide site at each one of the vertices of this face. In Figure 2 the normalized silver ion density distribution for PIM1 in one of these faces is shown. This density distribution has been calculated by averaging the positions of the silver ions within a plate of thickness a/16, a ) 5.075 Å being the lattice parameter.5 Since the density maps for the other two models are very similar, we do not present them. There are higher-density occupancy regions around the four tetrahedral sites and, to a lesser extent, around trigonal and octahedral sites (see the caption of Figure 2). This result is in agreement with the experimental work of Cava et al.,4 and our results are consistent with previous studies (see ref 1), which concluded that the diffusive motion of silver ions consists of jumps between adjacent tetrahedral sites via trigonal sites, with a small occupation of octahedral sites. We have also calculated the partial radial distribution functions, gab(r). gAgI and gII do not present significant differences from one model to another. The positions of the gII peaks correspond to the distances between the sites of a bcc lattice with lattice parameter a. The I-I coordination number (namely, the average number of iodides that are at distances smaller than that of the first gII minimum from a given iodide) is around 14, as in all bcc structures at high temperatures (8 nearest and 6 next-nearest neighbors). The first peak of gAgI is at 2.7 Å, slightly smaller than the distance between an iodide site and a tetrahedral site (2.8 Å). On the other hand, the gAgAg shown in Figure 3 presents interesting differences. The most striking one is that the first peak of the gAgAg for PIM1 is shifted backward at 3.2 Å. This shift, which has also been observed in the liquid phase (see below) and for the polarizable model of molten AgBr,23 can be

Polarization Effects on AgI

J. Phys. Chem. B, Vol. 112, No. 6, 2008 1723

Figure 3. Cation-cation radial distribution function, gAgAg(r), from MD simulations of R-AgI at 573 K using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line).

attributed to the screening of the repulsion between cations due to polarization. The negative ends of the anion dipoles attract the positive ions, and, therefore, the separation between the cations can be smaller than would be the case if the anions were not polarized. As it was already suggested by Gray-Weale and Madden,43 the PIM1 second peak (or shoulder) at 6.10 Å can be due to those pairs of neighboring cations attracted by different anion dipoles. Then, these two peaks can be interpreted as the result of the splitting of the RIM first peak at 4.3 Å because of the inclusion of polarization. In the PIM2s case, although the position of the first peak is very close to that for RIM because of the short-range damping effects, there is a shoulder at about the position of the PIM1 second peak. The shift of the gAgAg first peak found for PIM1 is similar to that found by Chahid and McGreevy for the superionic R-phase of CuI at high temperature and high pressure.44 The gCuCu(r) that they obtained from reverse Monte Carlo (RMC) modeling of experimental scattering data exhibits a first peak at a lower r than that found in MD simulations using a rigid ion pair potential. It is also interesting to relate gAgAg(r) to the occupation of the tetrahedral sites of the bcc cubic unit cells. In the three cases, there is a zero probability of finding silver ions closer together than 1.8 Å, which excludes the simultaneous occupation of two nearest tetrahedral sites. The position of the first peak of gAgAg for RIM and PIM2s (4.3 Å) is approximately the distance between two tetrahedral sites linked by the vector (1/2,1/2,1/ 2)a. From all possible distributions of two cations per unit bcc cell at the tetrahedral sites, this is the maximum nearest-neighbor separation. However, the first peak for PIM1 is at 3.2 Å, which means that two nearest cations tend to locate around tetrahedral sites linked by (1/2,1/4,1/4)a (or equivalent vectors of the same modulus). The RMC analysis performed by Nield et al.5 and the first-principles MD calculations carried out by Wood and Marzari45 also exhibited the first peak of gAgAg at around 3 Å. Nevertheless, they obtained the second peak at approximately the same position as the first peak of our RIM and PIM2s gAgAg. Their results, then, show the closer approach between neighboring silver ions found for PIM1, but also include the correlations at ∼4.3 Å found for RIM and PIM2s. 4.2. Liquid Structure. Figure 4 shows the partial radial distribution functions obtained from liquid-state simulations. Some characteristics are common to the three models: (a) each ion is surrounded by a shell of unlike nearest-neighbor ions; (b) gAgI oscillates in antiphase with gII; and (c) the silver ions, because of their smaller size, show a less marked structure and

Figure 4. Partial radial distribution functions, gAgAg(r) (solid line), gII(r) (dotted line), and gAgI(r) (dashed line), from MD simulations of molten AgI at 923 K using the RIM (top), PIM1 (middle), and PIM2s (bottom). The insets show gAgAg(r) in more detail.

a deeper penetration into the first coordination shell of a like ion. However, while the gAgAg for RIM oscillates in phase with gII, the first peak of gAgAg for PIM1 and PIM2s lies between the peaks of gAgI and gII, and the oscillations beyond it are strongly smoothed. For PIM1 and PIM2s, since the induced dipoles on the iodides screen the repulsion between Ag+ ions, the neighboring cations can approach at shorter distances, as it has been found for the R-phase. The PIM1 gAgAg for the liquid follows the same trends as that obtained for the R-phase. It has the first peak at a shorter distance (3.45 Å) than that for RIM (4.35 Å), and presents a small second peak at 5.85 Å (see the inset in Figure 4). Polarization effects are less intense for PIM2s because of the short-range damping, and the first peak of the gAgAg is in between those for RIM and PIM1, at 4.10 Å, with a shoulder at a position close to that of the second PIM1 peak (see the inset). It is worth noting that the first peak of the gAgAg for liquid PIM2s is at a shorter distance than that for the R-phase. The gII for the three models are almost identical. The iodides, because of their large size, are so close-packed that their structure does not depend significantly on the details of the pair potential or on the induced polarization effects. Nevertheless, comparison between the three gII shows little differences. The first peak of gII for PIM1 is slightly higher (2.49) and at a larger position (4.40 Å) than that for RIM (2.45 and 4.35 Å), with that for PIM2s in between. These little differences are probably related to the fact that the pair potential φ0II(r) used for PIM1 and PIM2s does not include the attractive contribution -PII/r4 added in φVR II (r) for RIM. Therefore, the repulsion between iodides is slightly weaker for RIM, and they are slightly less packed. It is as if the effective iodide size is slightly smaller for RIM. Although the gAgI for the three models are very similar, they present the following differences. The first peak of the gAgI for

1724 J. Phys. Chem. B, Vol. 112, No. 6, 2008 RIM is at 2.65 Å, and it is higher than those for PIM1 at 2.70 Å and PIM2s at 2.85 Å. These differences strengthen the idea that effective iodide size is slightly smaller for RIM. The inclusion of the effective monopole-induced dipole attraction term -PAgI/r4 in the pair potential φVR AgI(r) for RIM shifts its minimum at a shorter distance (2.46 Å) than that of φ0AgI(r) at 2.56 Å. The attractive force between the iodide dipoles and silver cations is also included in the polarizable models via the manybody interactions added to φ0AgI(r). However, despite the fact that the potential energy φPIM AgI (r) of two isolated unlike ions for the PIM1 case is equal to φVR AgI(r), it appears that the reduction of the effective iodide size due to the changes on the pair potentials is slightly more pronounced than that due to the induced many-body interactions. Some of the structural features found for PIM1 and PIM2s have also been observed in experimental studies of molten copper halides. The absence of a marked cation structure in molten CuCl was determined by Eisenberg et al.46 from experiments of neutron diffraction with isotopic substitution. In their RMC study on molten CuBr, which also melts from a superionic phase, Nield et al.47 found the peak of gCuCu at a lower r than that obtained from MD simulations using a rigid ion pair potential. They attributed this shift to polarization effects. Shirakawa et al.48 deduced the partial radial distribution functions of molten CuX (X ) Cl, Br, and I) from neutron diffraction data, and they found that, for the three cases, the gCuCu first peaks lie between those of gCuX and gXX. The partial structure factors Sab(k) are shown in Figure 5. The charge ordering is reflected in the reciprocal space by the fact that the main peaks of SAgAg and SII, and the valley of SAgI, are at the same wavenumber kM ≈ 1.7 Å-1. As expected from the gab(r), there are not appreciable differences in the three SII, and only slight ones in SAgI. There is a little displacement of the first minimum and the first maximum of SAgI, with the shortest k being for the PIM2s and the longest being for the RIM, that is, the inverse order of the gAgI peaks. The most salient differences are those exhibited by SAgAg. The main peak at kM of the SAgAg for PIM1 is much lower than that for PIM2s, which in turn is lower than that for RIM. The SAgAg for PIM1 presents a second peak at 2.25 Å-1 related to the second peak of the corresponding gAgAg at 5.85 Å. The distance between this second peak and the third one at 8.65 Å of the gAgAg for PIM1 (see Figure 4) is 2.80 Å, whose associated wavelength is (2π/2.80 Å) ) 2.24 Å-1. The SAgAg for PIM2s presents intermediate features between those for RIM and PIM1. It also exhibits a second maximum after the principal peak, but lower than that for PIM1. Nevertheless, the most striking polarization effect is the prepeak exhibited at k1 ≈ 1 Å-1 by the SAgAg for PIM1. In ref 24 we discussed its structural origin and suggested that it is the signature of a periodicity of low cation density zones (or voids) due to anion polarization effects. The shift to lower r of the gAgAg first peak for PIM1 with respect to RIM implies the existence of regions of high cation density. As a consequence of this inhomogeneous Ag+ distribution, the voids open up, giving rise to a new intermediate-range characteristic length scale, signaled by the prepeak. While this prepeak was not found in previous HNC calculations or MD simulations of molten AgI using pairwise potentials,15,17,19 it has been obtained in a recent ab initio MD study by Shimojo et al.25 In Figure 5a we compare the SAgAg for PIM1 with that by Shimojo et al., as well as that obtained by Kawakita et al.49 after RMC analysis of neutron diffraction data. Our MD SAgAg is similar to that by Shimojo et al. However, comparison between the corresponding gAgAg

Bitria´n and Trulla`s

Figure 5. Ashcroft-Langreth partial structure factors SAgAg(k) (b), SII(k) (c), and SAgI(k) (d), from MD simulations of molten AgI at 923 K using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line). The insets show the long-wavelength behavior. In panel a, the SAgAg(k) for PIM1 is compared with those by Shimojo et al.25 (open circles) and Kawakita et al.49 (solid circles).

shows some differences similar to those between our R-phase gAgAg and those by Nield et al.5 and by Wood and Marzari.45 The long-wavelength behavior of the partial structure factors is related to the degree of compactness of the ionic packing, and their limit values Sab(0) are proportional to the isothermal compressibility χT.40,50 For monohalides, the limit value is the same for the three partials, Sab(0) ) 0.5FNkBTχT, where kB is the Boltzmann’s constant, and T is the temperature. As can be seen in the insets of Figure 5, at the lowest k accessible in our MD simulations, the Sab(k) for PIM1 and PIM2s go to the same value, which is lower than that for RIM. The RIM compressibility is higher because the corresponding pair potential 4 φVR ab (r) includes the attractive contribution -Pab/r and, as it has been discussed above, the effective size of the RIM iodides is slightly smaller, and they are less closely packed, than in the polarizable models. This conjecture has been confirmed by a complementary MD simulation of a rigid ion model with φ0ab(r). This simulation shows that the corresponding Sab(k) go to the same limit value as those for PIM1 and PIM2s. It is worth noting that the values of SII and SAgI at the lowest wave numbers are close to the long-wavelength limit, while the values of SAgAg are higher and decay as k approaches zero. We interpret this result as follows. Because the silver ions are smaller than

Polarization Effects on AgI

J. Phys. Chem. B, Vol. 112, No. 6, 2008 1725

Figure 7. Charge static structure factors, SZZ(k), from MD simulations of molten AgI at 923 K using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line). The inset shows the long-wavelength behavior. The thin lines in the inset are the theoretical k f 0 behavior for the three models. Figure 6. Coherent static structure factor S(k) of molten AgI at 923 K from neutron scattering data (open circles),36 and MD simulations using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line).

iodides, they are less packed. This effect is reinforced for PIM1 and PIM2s since the induced polarization favors a less marked silver ion structure. The prepeak in SAgAg(k) leads to a prepeak in the coherent static structure factor S(k), which has also been observed experimentally by the Kyushu University group.36,49 Neutron diffraction data by Shirakawa et al.48 for molten copper halides also show a similar feature at ∼1 Å-1. In Figure 6 we compare the S(k) for the three models and that from neutron diffraction data. The three MD S(k) present a relatively structureless form as a result of subtle cancellations between the Sab(k). They present a broad main peak between 1.7 Å-1 and 3.0 Å-1 with the maximum at an intermediate k between the second peak of SII and the first SAgI maximum beyond the first valley. The maxima for the three models are slightly displaced toward a higher k (2.9 Å-1) with respect to the scattering data, and the oscillations beyond the main peak damp faster than those of the experiment. The PIM1 results are the closest to experiment, with the maximum value in agreement, although the mid shoulder at 2.25 Å-1, related to the second peak of SAgAg discussed above, is not present in the neutron scattering data. While the S(k) for RIM and PIM1 exhibits a small peak at kM ≈ 1.7 Å-1, where experimental data show a shoulder, the first maximum of the S(k) for PIM2s is at 2.1 Å-1, the same k of the PIM2s SAgAg second maximum. Since the values of coherent scattering lengths bAg and bI are not very different, S(k) resembles SNN(k) ) [S++(k) + S- -(k) + 2S+ -(k)]/2 and, thus, provides information about the topological order of the melt, namely, the correlations between ions irrespectively of their species. The chemical order is described by the charge static structure factor SZZ(k) ) [S++(k) + S- -(k) - 2S+ -(k)]/2. The results for the three models are shown in Figure 7. The pronounced main peak at kM ≈ 1.7 Å-1, less pronounced for PIM1, is a clear manifestation of the charge ordering. As it can be seen in the inset of Figure 7, the RIM SZZ(k) at k < 0.4 Å-1 practically fits the theoretical longwavelength limit approximation derived for rigid ion models, i.e., SZZ(k f 0) ) k2/kD2, where kD ) (4π z2e2FN/kBT)1/2 is the Debye wave number with z ) z+ ) |z-|.40,51 However, the corresponding values for PIM1 and PIM2s are higher than those for RIM, with those for PIM1 increasing much faster because its SAgAg(k) goes to its shoulder at about 1 Å-1. Recently we derived theoretically that the long-wavelength limit for ionic systems with point dipoles, as PIM1 and PIM2s, is given by SZZ(k f 0) ) [1 - C(k)]k2/kD2, where C(k) is related to spatial

TABLE 2: MD Results of the Self-Diffusion Coefficients, Da, and the Ionic Conductivity, σ, for r-AgI at 573 K and Molten AgI at 923 Ka R-AgI (573 K) DAg/10-5 cm2/s DI/10-5 cm2/s σ/(Ωcm)-1 σNE/(Ωcm)-1 ∆

molten AgI (923 K)

RIM

PIM2s

PIM1

RIM

PIM2s

PIM1

2.34 0 2.2 1.2 -0.87

2.70 0 2.1 1.3 -0.53

3.02 0 1.9 1.5 -0.27

6.50 2.22 3.4 2.5 -0.36

7.24 1.88 2.8 2.6 -0.08

9.08 1.75 2.8 3.1 0.10

a Experimental σ values: σ(573 K) ) 1.97 (Ω·cm)-1 (ref 8), and σ(923 K) ) 2.51 (Ω·cm)-1 (ref 37). Experimental DAg value: DAg(573 K) ) 2.41 × 10-5 cm2/s (ref 53).

correlations between charge and dipole moment densities.52 As can be seen in the inset of Figure 7, the three models fulfill their theoretical k f 0 limits derived in ref 52. 4.3. Ionic Transport Properties. In Table 2, we present the MD results of the self-diffusion coefficients, Da, and the ionic conductivities, σ, for R-AgI at 573 K and molten AgI at 923 K. In the superionic R-phase, the iodides remain around their lattice sites, and thus DI ) 0, whereas the silver ions show a liquidlike behavior and diffuse through the essentially rigid framework formed by the iodides. The three models retain in the liquid phase the superionic character of R-AgI, as the diffusivity of the silver ions is significantly larger than that for the iodides. Since silver ions are smaller than iodides, they can diffuse through the packed structure of the slowly diffusing iodides. The induced polarization effects are clearly observed on the silver ion diffusivity for both the liquid and R-phase. The value of DAg for PIM1 is larger than that for RIM, with that for PIM2s in between. The induced polarization on the iodides allows the small silver ions to move more easily through the interstices of the iodides structure, that is, it increases the free space for the cations that can diffuse through “faster” channels. On the contrary, the value of DI for liquid PIM1 is lower than that for RIM, with that for PIM2s in between. Although it could be a polarization effect, we believe that this result is mainly related to the small changes of the effective iodide size discussed above in going from the pair potential φVR ab (r) used for RIM to the φ0ab(r) for the polarizable ion models. Anyway, the DI in the liquid is similar in the three models. Unfortunately, we are not aware of experimental data for the self-diffusion coefficients, except the DAg in the R-phase given by Kvist and Tarneberg,53 which is near the RIM value (see Table 2). The normalized self-velocity autocorrelation functions, Ca(t), plotted in Figure 8 for the molten phase, help one to understand the ionic diffusion discussed above. The CAg(t) for the three models shows a typical diffusive behavior, with a slow decay and a weak minimum, in contrast with the oscillatory

1726 J. Phys. Chem. B, Vol. 112, No. 6, 2008

Bitria´n and Trulla`s

σ)

Figure 8. Normalized self-velocity autocorrelation functions, CAg(t) (top) and CI(t) (bottom), from MD simulations of molten AgI at 923 K, using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line).

behavior of the CI(t). As it was suggested in ref 54 for copper and silver halide melts, while the short-time microscopic motion of the small cations is solely determined by the first neighboring shell of unlike ions, the large anions are so closely packed that they experience a rattling motion within a double cage formed by unlike and like first neighbors. Since the induced polarization on the anions increases the free space for the cations, the CAg(t) initial decay is considerably slower for the two polarizable ion models than for the RIM, and the weak minimum is at longer t. The short-range damping polarization effects partially cancel this effect for PIM2s, whose CAg(t) has a minimum deeper than that for RIM. The CI(t) for the three models exhibit a pronounced backscattering minimum at about the same value of t, and subsequent damped oscillations with approximately the same frequency. The variation in the depth of the minimum is related to the differences between the 0 pair potentials φVR ab (r) and φab(r). Since the absence of the Pab 0 terms in φab(r) increases the short-range repulsion between anions, the backscattering is more pronounced for the polarizable models, consistently with the conclusions drawn in ref 54. Moreover, the reduction of the PIM1 minimum depth with respect to that for PIM2s is attributable to the non-damped polarization effects in the former model. The Ca(t) obtained from the simulations of the R-phase are qualitatively similar to those of the liquid, although the minimum of CAg(t) is deeper, and at a slightly shorter t, in the solid case, and the oscillations of CI(t) are considerably more pronounced. From the charge-density current autocorrelation function,

1

1

N

∑

ΛZZ(t) ) N〈vZ(t)‚vZ(0)〉 with vZ ) zivi 3 N i)1

(19)

and the mean square displacement of the charge-center position,

1 1 N ziri QZZ(t) ) N〈|rZ(t) - rZ(0)|2〉 with rZ ) 3 N i)1

∑

(20)

where we assume |zi| ) 1, we have evaluated the ionic conductivity σ by the corresponding Kubo and Einstein-like relations,

FNe2 kBT

F e2

∫0∞ ΛZZ(t)dt and σ ) kNBT lim tf∞

QZZ(t) 2t

(21)

The estimated uncertainty for the σ values reported in Table 2 is less than 5%, but greater than those for Da (less than 1%) because σ is a collective property that requires to be averaged over long simulations. In Figure 9 the time evolution of σ(t) ) (FNe2/kBT)∫t0ΛZZ(t′)dt′ obtained for the three models at the molten phase can be seen, along with its oscillating convergence at times longer than 2 ps. The uncertainty has been estimated from the fluctuating values of σ(t). Some features are common for both the R- and molten phases results. The conductivity for polarizable ion models is smaller than that for RIM, and in better agreement with experiment (see Table 2). In Figure 10 it can be seen that the QZZ(t) slopes for PIM1 and PIM2s at the molten phase are very similar, and clearly lower than that for RIM, in accordance with the converging values of σ(t). The larger silver ion diffusivity for PIM1 does not lead to a larger σ as would be expected from the Nernst-Einstein approximation, which for monohalides can be written as σNE ) FNe2(DAg+DI)/(2kBT).51 On the contrary, the PIM1 conductivity has the lowest value in both phases in spite of presenting the highest silver ion diffusivity. σ is not equal to σNE because, besides the self-diffusion term σNE, there is a contribution -∆σNE due to the correlations between distinct ions, i.e., σ ) σNE(1 - ∆). In all simulated cases, except molten PIM1, σ is higher than σNE and, thus, ∆ < 0. Moreover, the values of σ for the three models in each phase are ordered from lower to higher in the inverse order of σNE (and ∆), showing the different role played by distinct correlations. From the experimental values of σ (1.97 (Ω·cm)-1)8 and DAg (2.41 × 10-5 cm2/s)53 for R-ΑgI at 573 K, it is derived that ∆ ) -0.6, which is closer to the PIM2s result. We are not aware of experimental data of DAg, and thus of ∆, for molten AgI. The only experimental result on ∆ that we know for a superionic melt is ∆ ) 0.2 for molten CuCl,55 which is positive as for the PIM1 result. We recall that, in molten alkali halides, the experimental values of ∆ are also positive.56 Since the distinct contribution to ΛZZ(t) for AgI can be written as

ΛdZZ(t) ) ΛZZ(t) - 0.5[ΛAg(t) + ΛI(t)]

(22)

where ΛAg(t) and ΛI(t) are the (not normalized) self-velocity autocorrelation functions, the parameter ∆ is proportional to the area under ΛdZZ(t),

∆)-

2 DAg + DI

∫0∞ ΛdZZ(t)dt

(23)

In Figure 11 we present δ(t) ) ΛdZZ(t)/ΛZZ(0) obtained from the MD simulations of the molten phase. While the δ(t) for RIM is negative between 0 and 0.2 ps and positive between 0.2 and 0.7 ps, those for the two polarizable ion models show the same qualitative behavior with a negative minimum at about 0.33 ps and a positive maximum at around 0.58 ps. Nevertheless, ∆ is positive as the area under δ(t) is negative, and σ < σNE, for PIM1, whereas ∆ is negative and σ > σNE for PIM2s. 5. Concluding Remarks In this work we have carried out MD simulations for three models of AgI, an RIM with the effective interionic pair potential of the Vashishta and Rahman form and the param-

Polarization Effects on AgI

Figure 9. Values of the Kubo integral for the conductivity as a function of the upper limit time, σ(t) ) (FNe2/kBT)∫t0ΛZZ(t′)dt′, from MD simulations of molten AgI at 923 K, using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line).

Figure 10. Mean square displacement of the charge-center position, QZZ(t), from MD simulations of molten AgI at 923 K, using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line).

Figure 11. Normalized distinct velocity correlation function δ(t) from MD simulations of molten AgI at 923 K using the RIM (dotted line), PIM1 (solid line), and PIM2s (dashed line).

etrization proposed by Shimojo,19 and two polarizable ion models (PIM1 and PIM2s). The latter are based on the RIM pair potential, to which the polarization effects are added. In PIM1, only the iodides are assumed to be polarizable, and the dipole moment of each iodide is induced by the local electric field due to all other ions. In PIM2s both species are polarizable, and the dipoles are not only induced by the field but also by the short-range overlap effects due to the neighbors. In order to guarantee the stability of the simulations, the short-range damping polarization must oppose the dipole contributions of the field created by both the charges and dipoles of other ions. Under these conditions, we solve the “polarization catastrophe” problem, which had prevented the simulation of a similar model in our previous work on AgBr.23 The solid R-phase of AgI has been simulated at 573 K. The three models reproduce the superionic character of R-AgI. The averaged spatial distribution of the diffusive silver ions within the bcc iodide sublattice is qualitatively similar in the three models, and reproduces the well-known experimental results. The cations tend to be located around the tetrahedral sites and, to a lesser extent, around trigonal and octahedral sites. However, while, in the RIM and PIM2s, the separation between neighboring cations is maximized (4.3 Å), the PIM1 gAgAg peaks at 3.2

J. Phys. Chem. B, Vol. 112, No. 6, 2008 1727 Å, which means that in this model two nearest silver ions tend to locate around tetrahedral sites linked by (1/2,1/4,1/4)a or equivalent vectors of the same modulus. The RMC analysis by Nield et al.5 and the ab initio MD calculations by Wood and Marzari45 agree on the existence of a peak at around 3 Å. Regarding the liquid, we have found that the iodide structure does not depend significantly on the polarization effects. On the contrary, it is mainly the silver ions that are affected, which can be attributed to the screening of the cations repulsion due to the induced dipoles on the iodides. The silver ion structure for the two polarizable models is less marked, and the cations penetration within the first shell is deeper, than for the RIM. Furthermore, as in the solid-state simulations, the distance between nearest cations is shorter for PIM1, and to a lesser extent for PIM2s, with respect to RIM. As a result of the induced polarization, the silver ions have more free space to move, and tend to diffuse through the interstices of the iodides structure, in such a way that the silver ions’ self-diffusion coefficients in both the R- and molten phases are higher for PIM1 than for PIM2s and RIM (in this order). In spite of this fact, and due to the distinct velocity correlations, the ionic conductivity is lower for the polarizable models than that for RIM. The superionic character of the melt in the three models is reflected in the significantly larger mobility of the cations compared to that of anions. The static coherent structure factor S(k) has been calculated for the liquid and compared to experimental data. The three models present a broad main peak between 1.7 Å-1 and 3.0 Å-1, similar to that from neutron scattering experiments, but the PIM1 S(k) is the only one that reproduces the experimental prepeak at about 1 Å-1. The prepeak is exhibited also by the PIM1 Ag-Ag partial structure factor. We have suggested that it is due to the inhomogeneous distribution in space of silver ions, as a consequence of polarization effects.24 The closer approach between neighboring silver ions observed in PIM1 causes the appearance of low Ag+ density zones, whose periodicity gives rise to the prepeak. Moreover, the PIM1 and PIM2s conductivity is in better agreement with experimental values, and the PIM1 deviation parameter ∆ is positive for molten AgI, like the experimental results for molten alkali halides and molten CuCl. The overall results for PIM1 present a better agreement with experimental data than the other two models. In particular, the appearance of the prepeak in S(k) leads us to conclude that the anionic induced polarization plays an important role in the real system. However, the point dipole approximation of PIM1 is not the last word on modeling AgI, and more improvements are necessary to gain further insight into this fascinating material. The improvements may be achieved, for instance, by using more refined pair potential contributions than those of this work, assuming more sophisticated damping polarization interactions that cancel the polarization effects to a lesser extent than in PIM2s, or adding quadrupole polarization effects. Besides the empirical model approximation, the ab initio simulation techniques can help to reach a better understanding of AgI. Acknowledgment. This work was supported by DGI of Spain (Grant No. FIS2006-12436-C02-01), the DURSI of the Generalitat of Catalonia (Grant No. 2005SGR-00779), and European Union FEDER funds (Grant No. UNPC-E015). V.B. also thanks the Spanish Ministry of Education and Science for FPU Grant No. AP2003-3408. References and Notes (1) Keen, D. A. J. Phys.: Condens. Matter 2002, 14, R819. (2) Burley, G. J. Phys. Chem. 1964, 68, 1111.

1728 J. Phys. Chem. B, Vol. 112, No. 6, 2008 (3) Hoshino, S.; Sakuma, T.; Fujii, Y. Solid State Commun. 1977, 22, 763. (4) Cava, R. J.; Reidinger, F.; Wuensch, B. J. Solid State Commun. 1977, 24, 411. (5) Nield, V. M.; Keen, D. A.; Hayes, W.; McGreevy, R. L. Solid State Ionics 1993, 66, 247. (6) Cava, R. J.; Fleming, R. M.; Rietman, E. A. Solid State Ionics 1983, 9-10, 1347. (7) Boyce, J. B.; Hayes, T. M.; Stutius, W.; Mikkelsen, J. C., Jr. Phys. ReV. Lett. 1977, 38, 1362. (8) Tubandt, C.; Lorenz, E. Z. Phys. Chem. 1914, 87, 513. (9) Vashishta, P.; Rahman, A. Phys. ReV. Lett. 1978, 40, 1337. (10) Parrinello, M.; Rahman, A.; Vashishta, P. Phys. ReV. Lett. 1983, 50, 1073. (11) Tallon, J. L. Phys. ReV. B 1988, 38, 9069. (12) Chiarotti, G. L.; Jacucci, G.; Rahman, A. Phys. ReV. Lett. 1986, 57, 2395. (13) Madden, P. A.; O’Sullivan, K. F.; Chiarotti, G. Phys. ReV. B 1992, 45, 10206. (14) Howe, M. A.; McGreevy, R. L.; Mitchell, E. W. J. Z. Phys. B 1985, 62, 15. (15) Stafford, A. J.; Silbert, M. Z. Phys. B 1987, 67, 31. (16) Takahashi, H.; Takeda, S.; Harada, S.; Tamaki, S. J. Phys. Soc. Jpn. 1988, 57, 562. (17) Stafford, A. J.; Silbert, M.; Trulla`s, J.; Giro´, A. J. Phys.: Condens. Matter 1990, 2, 6631. (18) Trulla`s, J.; Giro´, A.; Silbert, M. J. Phys.: Condens. Matter 1990, 2, 6643. (19) Shimojo, F.; Kobayashi, M. J. Phys. Soc. Jpn. 1991, 60, 3725. (20) Madden, P. A.; Wilson, M. Chem. Soc. ReV. 1996, 25, 339. (21) Wilson, M.; Madden, P. A.; Costa-Cabral, B. J. J. Phys. Chem. 1996, 100, 1227. (22) Trulla`s, J.; Alcaraz, O.; Gonza´lez, L. E.; Silbert, M. J. Phys. Chem. B 2003, 107, 282. (23) Bitria´n, V.; Trulla`s, J. J. Phys. Chem. B 2006, 110, 7490. (24) Bitria´n, V.; Trulla`s, J.; Silbert, M. J. Chem. Phys. 2007, 126, 021105. (25) Shimojo, F.; Inoue, T.; Aniya, M.; Sugahara, T.; Miyata, Y. J. Phys. Soc. Jpn. 2006, 75, 114602. (26) Wilson, M.; Madden, P. A. J. Phys.: Condens. Matter 1993, 5, 2687. (27) Rick, S. W.; Stuart, S. J. In ReViews in Computational Chemistry; Lipkowitz, K. B., Boyd, D. B., Eds.; John Wiley and Sons: New York, 2002; Vol. 18. (28) Hutchinson, F.; Wilson, M.; Madden, P. A. Mol. Phys. 2001, 99, 811. Wilson, M.; Madden, P. A. J. Phys.: Condens. Matter 1994, 6, 159.

Bitria´n and Trulla`s (29) Thole, B. T. Chem. Phys. 1981, 59, 341. (30) Burnham, C. J.; Li, J.; Xantheas, S. S.; Leslie, M. J. Chem. Phys. 1999, 110, 4566. (31) Tang, K. T.; Toennies, J. P. J. Chem. Phys. 1984, 80, 3726. (32) Ahlstro¨m, P.; Wallqvist, A.; Engstro¨m, S.; Jo¨nsson, B. Mol. Phys. 1989, 68, 563. (33) Bo¨ttcher, C. J. F. Theory of Electric Polarization, 2nd ed.; Elsevier: Amsterdam, 1993. (34) Bucher, M. Phys. ReV. B 1984, 30, 947. (35) Alcaraz, O.; Bitria´n, V.; Trulla`s, J. J. Chem. Phys. 2007, 127, 154508. (36) Bitria´n, V.; Trulla`s, J.; Silbert, M.; Enosaki, T.; Kawakita, Y.; Takeda, S. J. Chem. Phys. 2006, 125, 184510. (37) Janz, G. J.; Dampier, F. W.; Lakdhminaryanah, G. R.; Lorentz, P. K.; Tomkins, R. P. T. Molten Salts; National Bureau of Standards Reference Data Series; National Bureau of Standards: Washington, D.C., 1968; Vol. 15. (38) Vesely, F. J. J. Comput. Phys. 1977, 24, 361. (39) Ashcroft, N. W.; Langreth, D. C. Phys. ReV. 1967, 156, 685. (40) Rovere, M.; Tosi, M. P. Rep. Prog. Phys. 1986, 49, 1001. (41) Sears, V. F. Neutron News 1992, 3, 26. (42) Tosi, M. P.; Price, D. L.; Saboungi, M.-L. Annu. ReV. Phys. Chem. 1993, 44, 173. (43) Gray-Weale, A.; Madden, P. A. Mol. Phys. 2003, 101, 1761. (44) Chahid, A.; McGreevy, R. L. J. Phys.: Condens. Matter 1998, 10, 2597. (45) Wood, B. C.; Marzari, N. Phys. ReV. Lett. 2006, 97, 166401. (46) Eisenberg, S.; Jal, J. F.; Dupuy, J.; Chieux, P.; Knoll, W. Phil. Mag. A 1982, 46, 195. (47) Nield, V. M.; McGreevy, R. L.; Keen, D. A.; Hayes, W. Physica B 1994, 202, 159. (48) Shirakawa, Y.; Saito, M.; Tamaki, S.; Inui, M.; Takeda, S. J. Phys. Soc. Jpn. 1991, 60, 2678. (49) Kawakita, Y.; Tahara, S.; Fujii, H.; Kohara, S.; Takeda, S. J. Phys.: Condens. Matter 2007, 19, 335201. (50) McGreevy, R. L. Solid State Phys. 1987, 40, 247. (51) Hansen, J. P.; McDonald, I. R. Theory of Simple Liquids, 2nd ed.; Academic: London, 1986. (52) Bitria´n, V.; Trulla`s, J. J. Phys: Conf. Ser., submitted for publication, 2007. (53) Kvist, A.; Tarneberg, R. Z. Naturforsch. 1970, 25A, 257. (54) Alcaraz, O.; Trulla`s, J. J. Chem. Phys. 2000, 113, 10635. (55) Poignet, J. C.; Barbier, M. J. Electrochim. Acta 1981, 26, 1429. (56) Ciccotti, G.; Jacucci, G.; McDonald, I. R. Phys. ReV. A 1976, 13, 426.